1 /* 2 * Copyright (c) 1982, 1986, 1989, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * sendfile(2) and related extensions: 6 * Copyright (c) 1998, David Greenman. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)uipc_syscalls.c 8.4 (Berkeley) 2/21/94 33 * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $ 34 */ 35 36 #include "opt_ktrace.h" 37 #include "opt_sctp.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/sysproto.h> 43 #include <sys/malloc.h> 44 #include <sys/filedesc.h> 45 #include <sys/event.h> 46 #include <sys/proc.h> 47 #include <sys/fcntl.h> 48 #include <sys/file.h> 49 #include <sys/filio.h> 50 #include <sys/kern_syscall.h> 51 #include <sys/mbuf.h> 52 #include <sys/protosw.h> 53 #include <sys/sfbuf.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/socketops.h> 57 #include <sys/uio.h> 58 #include <sys/vnode.h> 59 #include <sys/lock.h> 60 #include <sys/mount.h> 61 #ifdef KTRACE 62 #include <sys/ktrace.h> 63 #endif 64 #include <vm/vm.h> 65 #include <vm/vm_object.h> 66 #include <vm/vm_page.h> 67 #include <vm/vm_pageout.h> 68 #include <vm/vm_kern.h> 69 #include <vm/vm_extern.h> 70 #include <sys/file2.h> 71 #include <sys/signalvar.h> 72 #include <sys/serialize.h> 73 74 #include <sys/thread2.h> 75 #include <sys/msgport2.h> 76 #include <sys/socketvar2.h> 77 #include <net/netmsg2.h> 78 #include <vm/vm_page2.h> 79 80 #ifdef SCTP 81 #include <netinet/sctp_peeloff.h> 82 #endif /* SCTP */ 83 84 extern int use_soaccept_pred_fast; 85 extern int use_sendfile_async; 86 extern int use_soconnect_async; 87 88 /* 89 * System call interface to the socket abstraction. 90 */ 91 92 extern struct fileops socketops; 93 94 /* 95 * socket_args(int domain, int type, int protocol) 96 */ 97 int 98 kern_socket(int domain, int type, int protocol, int *res) 99 { 100 struct thread *td = curthread; 101 struct filedesc *fdp = td->td_proc->p_fd; 102 struct socket *so; 103 struct file *fp; 104 int fd, error; 105 106 KKASSERT(td->td_lwp); 107 108 error = falloc(td->td_lwp, &fp, &fd); 109 if (error) 110 return (error); 111 error = socreate(domain, &so, type, protocol, td); 112 if (error) { 113 fsetfd(fdp, NULL, fd); 114 } else { 115 fp->f_type = DTYPE_SOCKET; 116 fp->f_flag = FREAD | FWRITE; 117 fp->f_ops = &socketops; 118 fp->f_data = so; 119 *res = fd; 120 fsetfd(fdp, fp, fd); 121 } 122 fdrop(fp); 123 return (error); 124 } 125 126 /* 127 * MPALMOSTSAFE 128 */ 129 int 130 sys_socket(struct socket_args *uap) 131 { 132 int error; 133 134 error = kern_socket(uap->domain, uap->type, uap->protocol, 135 &uap->sysmsg_iresult); 136 137 return (error); 138 } 139 140 int 141 kern_bind(int s, struct sockaddr *sa) 142 { 143 struct thread *td = curthread; 144 struct proc *p = td->td_proc; 145 struct file *fp; 146 int error; 147 148 KKASSERT(p); 149 error = holdsock(p->p_fd, s, &fp); 150 if (error) 151 return (error); 152 error = sobind((struct socket *)fp->f_data, sa, td); 153 fdrop(fp); 154 return (error); 155 } 156 157 /* 158 * bind_args(int s, caddr_t name, int namelen) 159 * 160 * MPALMOSTSAFE 161 */ 162 int 163 sys_bind(struct bind_args *uap) 164 { 165 struct sockaddr *sa; 166 int error; 167 168 error = getsockaddr(&sa, uap->name, uap->namelen); 169 if (error) 170 return (error); 171 error = kern_bind(uap->s, sa); 172 kfree(sa, M_SONAME); 173 174 return (error); 175 } 176 177 int 178 kern_listen(int s, int backlog) 179 { 180 struct thread *td = curthread; 181 struct proc *p = td->td_proc; 182 struct file *fp; 183 int error; 184 185 KKASSERT(p); 186 error = holdsock(p->p_fd, s, &fp); 187 if (error) 188 return (error); 189 error = solisten((struct socket *)fp->f_data, backlog, td); 190 fdrop(fp); 191 return(error); 192 } 193 194 /* 195 * listen_args(int s, int backlog) 196 * 197 * MPALMOSTSAFE 198 */ 199 int 200 sys_listen(struct listen_args *uap) 201 { 202 int error; 203 204 error = kern_listen(uap->s, uap->backlog); 205 return (error); 206 } 207 208 /* 209 * Returns the accepted socket as well. 210 * 211 * NOTE! The sockets sitting on so_comp/so_incomp might have 0 refs, the 212 * pool token is absolutely required to avoid a sofree() race, 213 * as well as to avoid tailq handling races. 214 */ 215 static boolean_t 216 soaccept_predicate(struct netmsg_so_notify *msg) 217 { 218 struct socket *head = msg->base.nm_so; 219 struct socket *so; 220 221 if (head->so_error != 0) { 222 msg->base.lmsg.ms_error = head->so_error; 223 return (TRUE); 224 } 225 lwkt_getpooltoken(head); 226 if (!TAILQ_EMPTY(&head->so_comp)) { 227 /* Abuse nm_so field as copy in/copy out parameter. XXX JH */ 228 so = TAILQ_FIRST(&head->so_comp); 229 TAILQ_REMOVE(&head->so_comp, so, so_list); 230 head->so_qlen--; 231 soclrstate(so, SS_COMP); 232 so->so_head = NULL; 233 soreference(so); 234 235 lwkt_relpooltoken(head); 236 237 msg->base.lmsg.ms_error = 0; 238 msg->base.nm_so = so; 239 return (TRUE); 240 } 241 lwkt_relpooltoken(head); 242 if (head->so_state & SS_CANTRCVMORE) { 243 msg->base.lmsg.ms_error = ECONNABORTED; 244 return (TRUE); 245 } 246 if (msg->nm_fflags & FNONBLOCK) { 247 msg->base.lmsg.ms_error = EWOULDBLOCK; 248 return (TRUE); 249 } 250 251 return (FALSE); 252 } 253 254 /* 255 * The second argument to kern_accept() is a handle to a struct sockaddr. 256 * This allows kern_accept() to return a pointer to an allocated struct 257 * sockaddr which must be freed later with FREE(). The caller must 258 * initialize *name to NULL. 259 */ 260 int 261 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res) 262 { 263 struct thread *td = curthread; 264 struct filedesc *fdp = td->td_proc->p_fd; 265 struct file *lfp = NULL; 266 struct file *nfp = NULL; 267 struct sockaddr *sa; 268 struct socket *head, *so; 269 struct netmsg_so_notify msg; 270 int fd; 271 u_int fflag; /* type must match fp->f_flag */ 272 int error, tmp; 273 274 *res = -1; 275 if (name && namelen && *namelen < 0) 276 return (EINVAL); 277 278 error = holdsock(td->td_proc->p_fd, s, &lfp); 279 if (error) 280 return (error); 281 282 error = falloc(td->td_lwp, &nfp, &fd); 283 if (error) { /* Probably ran out of file descriptors. */ 284 fdrop(lfp); 285 return (error); 286 } 287 head = (struct socket *)lfp->f_data; 288 if ((head->so_options & SO_ACCEPTCONN) == 0) { 289 error = EINVAL; 290 goto done; 291 } 292 293 if (fflags & O_FBLOCKING) 294 fflags |= lfp->f_flag & ~FNONBLOCK; 295 else if (fflags & O_FNONBLOCKING) 296 fflags |= lfp->f_flag | FNONBLOCK; 297 else 298 fflags = lfp->f_flag; 299 300 if (use_soaccept_pred_fast) { 301 boolean_t pred; 302 303 /* Initialize necessary parts for soaccept_predicate() */ 304 netmsg_init(&msg.base, head, &netisr_apanic_rport, 0, NULL); 305 msg.nm_fflags = fflags; 306 307 lwkt_getpooltoken(head); 308 pred = soaccept_predicate(&msg); 309 lwkt_relpooltoken(head); 310 311 if (pred) { 312 error = msg.base.lmsg.ms_error; 313 if (error) 314 goto done; 315 else 316 goto accepted; 317 } 318 } 319 320 /* optimize for uniprocessor case later XXX JH */ 321 netmsg_init_abortable(&msg.base, head, &curthread->td_msgport, 322 0, netmsg_so_notify, netmsg_so_notify_doabort); 323 msg.nm_predicate = soaccept_predicate; 324 msg.nm_fflags = fflags; 325 msg.nm_etype = NM_REVENT; 326 error = lwkt_domsg(head->so_port, &msg.base.lmsg, PCATCH); 327 if (error) 328 goto done; 329 330 accepted: 331 /* 332 * At this point we have the connection that's ready to be accepted. 333 * 334 * NOTE! soaccept_predicate() ref'd so for us, and soaccept() expects 335 * to eat the ref and turn it into a descriptor. 336 */ 337 so = msg.base.nm_so; 338 339 fflag = lfp->f_flag; 340 341 /* connection has been removed from the listen queue */ 342 KNOTE(&head->so_rcv.ssb_kq.ki_note, 0); 343 344 if (head->so_sigio != NULL) 345 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 346 347 nfp->f_type = DTYPE_SOCKET; 348 nfp->f_flag = fflag; 349 nfp->f_ops = &socketops; 350 nfp->f_data = so; 351 /* Sync socket nonblocking/async state with file flags */ 352 tmp = fflag & FNONBLOCK; 353 fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL); 354 tmp = fflag & FASYNC; 355 fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL); 356 357 sa = NULL; 358 if (so->so_faddr != NULL) { 359 sa = so->so_faddr; 360 so->so_faddr = NULL; 361 362 soaccept_generic(so); 363 error = 0; 364 } else { 365 error = soaccept(so, &sa); 366 } 367 368 /* 369 * Set the returned name and namelen as applicable. Set the returned 370 * namelen to 0 for older code which might ignore the return value 371 * from accept. 372 */ 373 if (error == 0) { 374 if (sa && name && namelen) { 375 if (*namelen > sa->sa_len) 376 *namelen = sa->sa_len; 377 *name = sa; 378 } else { 379 if (sa) 380 kfree(sa, M_SONAME); 381 } 382 } 383 384 done: 385 /* 386 * If an error occured clear the reserved descriptor, else associate 387 * nfp with it. 388 * 389 * Note that *res is normally ignored if an error is returned but 390 * a syscall message will still have access to the result code. 391 */ 392 if (error) { 393 fsetfd(fdp, NULL, fd); 394 } else { 395 *res = fd; 396 fsetfd(fdp, nfp, fd); 397 } 398 fdrop(nfp); 399 fdrop(lfp); 400 return (error); 401 } 402 403 /* 404 * accept(int s, caddr_t name, int *anamelen) 405 * 406 * MPALMOSTSAFE 407 */ 408 int 409 sys_accept(struct accept_args *uap) 410 { 411 struct sockaddr *sa = NULL; 412 int sa_len; 413 int error; 414 415 if (uap->name) { 416 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 417 if (error) 418 return (error); 419 420 error = kern_accept(uap->s, 0, &sa, &sa_len, 421 &uap->sysmsg_iresult); 422 423 if (error == 0) 424 error = copyout(sa, uap->name, sa_len); 425 if (error == 0) { 426 error = copyout(&sa_len, uap->anamelen, 427 sizeof(*uap->anamelen)); 428 } 429 if (sa) 430 kfree(sa, M_SONAME); 431 } else { 432 error = kern_accept(uap->s, 0, NULL, 0, 433 &uap->sysmsg_iresult); 434 } 435 return (error); 436 } 437 438 /* 439 * extaccept(int s, int fflags, caddr_t name, int *anamelen) 440 * 441 * MPALMOSTSAFE 442 */ 443 int 444 sys_extaccept(struct extaccept_args *uap) 445 { 446 struct sockaddr *sa = NULL; 447 int sa_len; 448 int error; 449 int fflags = uap->flags & O_FMASK; 450 451 if (uap->name) { 452 error = copyin(uap->anamelen, &sa_len, sizeof(sa_len)); 453 if (error) 454 return (error); 455 456 error = kern_accept(uap->s, fflags, &sa, &sa_len, 457 &uap->sysmsg_iresult); 458 459 if (error == 0) 460 error = copyout(sa, uap->name, sa_len); 461 if (error == 0) { 462 error = copyout(&sa_len, uap->anamelen, 463 sizeof(*uap->anamelen)); 464 } 465 if (sa) 466 kfree(sa, M_SONAME); 467 } else { 468 error = kern_accept(uap->s, fflags, NULL, 0, 469 &uap->sysmsg_iresult); 470 } 471 return (error); 472 } 473 474 475 /* 476 * Returns TRUE if predicate satisfied. 477 */ 478 static boolean_t 479 soconnected_predicate(struct netmsg_so_notify *msg) 480 { 481 struct socket *so = msg->base.nm_so; 482 483 /* check predicate */ 484 if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) { 485 msg->base.lmsg.ms_error = so->so_error; 486 return (TRUE); 487 } 488 489 return (FALSE); 490 } 491 492 int 493 kern_connect(int s, int fflags, struct sockaddr *sa) 494 { 495 struct thread *td = curthread; 496 struct proc *p = td->td_proc; 497 struct file *fp; 498 struct socket *so; 499 int error, interrupted = 0; 500 501 error = holdsock(p->p_fd, s, &fp); 502 if (error) 503 return (error); 504 so = (struct socket *)fp->f_data; 505 506 if (fflags & O_FBLOCKING) 507 /* fflags &= ~FNONBLOCK; */; 508 else if (fflags & O_FNONBLOCKING) 509 fflags |= FNONBLOCK; 510 else 511 fflags = fp->f_flag; 512 513 if (so->so_state & SS_ISCONNECTING) { 514 error = EALREADY; 515 goto done; 516 } 517 error = soconnect(so, sa, td, use_soconnect_async ? FALSE : TRUE); 518 if (error) 519 goto bad; 520 if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) { 521 error = EINPROGRESS; 522 goto done; 523 } 524 if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { 525 struct netmsg_so_notify msg; 526 527 netmsg_init_abortable(&msg.base, so, 528 &curthread->td_msgport, 529 0, 530 netmsg_so_notify, 531 netmsg_so_notify_doabort); 532 msg.nm_predicate = soconnected_predicate; 533 msg.nm_etype = NM_REVENT; 534 error = lwkt_domsg(so->so_port, &msg.base.lmsg, PCATCH); 535 if (error == EINTR || error == ERESTART) 536 interrupted = 1; 537 } 538 if (error == 0) { 539 error = so->so_error; 540 so->so_error = 0; 541 } 542 bad: 543 if (!interrupted) 544 soclrstate(so, SS_ISCONNECTING); 545 if (error == ERESTART) 546 error = EINTR; 547 done: 548 fdrop(fp); 549 return (error); 550 } 551 552 /* 553 * connect_args(int s, caddr_t name, int namelen) 554 * 555 * MPALMOSTSAFE 556 */ 557 int 558 sys_connect(struct connect_args *uap) 559 { 560 struct sockaddr *sa; 561 int error; 562 563 error = getsockaddr(&sa, uap->name, uap->namelen); 564 if (error) 565 return (error); 566 error = kern_connect(uap->s, 0, sa); 567 kfree(sa, M_SONAME); 568 569 return (error); 570 } 571 572 /* 573 * connect_args(int s, int fflags, caddr_t name, int namelen) 574 * 575 * MPALMOSTSAFE 576 */ 577 int 578 sys_extconnect(struct extconnect_args *uap) 579 { 580 struct sockaddr *sa; 581 int error; 582 int fflags = uap->flags & O_FMASK; 583 584 error = getsockaddr(&sa, uap->name, uap->namelen); 585 if (error) 586 return (error); 587 error = kern_connect(uap->s, fflags, sa); 588 kfree(sa, M_SONAME); 589 590 return (error); 591 } 592 593 int 594 kern_socketpair(int domain, int type, int protocol, int *sv) 595 { 596 struct thread *td = curthread; 597 struct filedesc *fdp; 598 struct file *fp1, *fp2; 599 struct socket *so1, *so2; 600 int fd1, fd2, error; 601 602 fdp = td->td_proc->p_fd; 603 error = socreate(domain, &so1, type, protocol, td); 604 if (error) 605 return (error); 606 error = socreate(domain, &so2, type, protocol, td); 607 if (error) 608 goto free1; 609 error = falloc(td->td_lwp, &fp1, &fd1); 610 if (error) 611 goto free2; 612 sv[0] = fd1; 613 fp1->f_data = so1; 614 error = falloc(td->td_lwp, &fp2, &fd2); 615 if (error) 616 goto free3; 617 fp2->f_data = so2; 618 sv[1] = fd2; 619 error = soconnect2(so1, so2); 620 if (error) 621 goto free4; 622 if (type == SOCK_DGRAM) { 623 /* 624 * Datagram socket connection is asymmetric. 625 */ 626 error = soconnect2(so2, so1); 627 if (error) 628 goto free4; 629 } 630 fp1->f_type = fp2->f_type = DTYPE_SOCKET; 631 fp1->f_flag = fp2->f_flag = FREAD|FWRITE; 632 fp1->f_ops = fp2->f_ops = &socketops; 633 fsetfd(fdp, fp1, fd1); 634 fsetfd(fdp, fp2, fd2); 635 fdrop(fp1); 636 fdrop(fp2); 637 return (error); 638 free4: 639 fsetfd(fdp, NULL, fd2); 640 fdrop(fp2); 641 free3: 642 fsetfd(fdp, NULL, fd1); 643 fdrop(fp1); 644 free2: 645 (void)soclose(so2, 0); 646 free1: 647 (void)soclose(so1, 0); 648 return (error); 649 } 650 651 /* 652 * socketpair(int domain, int type, int protocol, int *rsv) 653 */ 654 int 655 sys_socketpair(struct socketpair_args *uap) 656 { 657 int error, sockv[2]; 658 659 error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv); 660 661 if (error == 0) { 662 error = copyout(sockv, uap->rsv, sizeof(sockv)); 663 664 if (error != 0) { 665 kern_close(sockv[0]); 666 kern_close(sockv[1]); 667 } 668 } 669 670 return (error); 671 } 672 673 int 674 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio, 675 struct mbuf *control, int flags, size_t *res) 676 { 677 struct thread *td = curthread; 678 struct lwp *lp = td->td_lwp; 679 struct proc *p = td->td_proc; 680 struct file *fp; 681 size_t len; 682 int error; 683 struct socket *so; 684 #ifdef KTRACE 685 struct iovec *ktriov = NULL; 686 struct uio ktruio; 687 #endif 688 689 error = holdsock(p->p_fd, s, &fp); 690 if (error) 691 return (error); 692 #ifdef KTRACE 693 if (KTRPOINT(td, KTR_GENIO)) { 694 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 695 696 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 697 bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen); 698 ktruio = *auio; 699 } 700 #endif 701 len = auio->uio_resid; 702 so = (struct socket *)fp->f_data; 703 if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 704 if (fp->f_flag & FNONBLOCK) 705 flags |= MSG_FNONBLOCKING; 706 } 707 error = so_pru_sosend(so, sa, auio, NULL, control, flags, td); 708 if (error) { 709 if (auio->uio_resid != len && (error == ERESTART || 710 error == EINTR || error == EWOULDBLOCK)) 711 error = 0; 712 if (error == EPIPE && !(flags & MSG_NOSIGNAL) && 713 !(so->so_options & SO_NOSIGPIPE)) 714 lwpsignal(p, lp, SIGPIPE); 715 } 716 #ifdef KTRACE 717 if (ktriov != NULL) { 718 if (error == 0) { 719 ktruio.uio_iov = ktriov; 720 ktruio.uio_resid = len - auio->uio_resid; 721 ktrgenio(lp, s, UIO_WRITE, &ktruio, error); 722 } 723 kfree(ktriov, M_TEMP); 724 } 725 #endif 726 if (error == 0) 727 *res = len - auio->uio_resid; 728 fdrop(fp); 729 return (error); 730 } 731 732 /* 733 * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen) 734 * 735 * MPALMOSTSAFE 736 */ 737 int 738 sys_sendto(struct sendto_args *uap) 739 { 740 struct thread *td = curthread; 741 struct uio auio; 742 struct iovec aiov; 743 struct sockaddr *sa = NULL; 744 int error; 745 746 if (uap->to) { 747 error = getsockaddr(&sa, uap->to, uap->tolen); 748 if (error) 749 return (error); 750 } 751 aiov.iov_base = uap->buf; 752 aiov.iov_len = uap->len; 753 auio.uio_iov = &aiov; 754 auio.uio_iovcnt = 1; 755 auio.uio_offset = 0; 756 auio.uio_resid = uap->len; 757 auio.uio_segflg = UIO_USERSPACE; 758 auio.uio_rw = UIO_WRITE; 759 auio.uio_td = td; 760 761 error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags, 762 &uap->sysmsg_szresult); 763 764 if (sa) 765 kfree(sa, M_SONAME); 766 return (error); 767 } 768 769 /* 770 * sendmsg_args(int s, caddr_t msg, int flags) 771 * 772 * MPALMOSTSAFE 773 */ 774 int 775 sys_sendmsg(struct sendmsg_args *uap) 776 { 777 struct thread *td = curthread; 778 struct msghdr msg; 779 struct uio auio; 780 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 781 struct sockaddr *sa = NULL; 782 struct mbuf *control = NULL; 783 int error; 784 785 error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg)); 786 if (error) 787 return (error); 788 789 /* 790 * Conditionally copyin msg.msg_name. 791 */ 792 if (msg.msg_name) { 793 error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen); 794 if (error) 795 return (error); 796 } 797 798 /* 799 * Populate auio. 800 */ 801 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 802 &auio.uio_resid); 803 if (error) 804 goto cleanup2; 805 auio.uio_iov = iov; 806 auio.uio_iovcnt = msg.msg_iovlen; 807 auio.uio_offset = 0; 808 auio.uio_segflg = UIO_USERSPACE; 809 auio.uio_rw = UIO_WRITE; 810 auio.uio_td = td; 811 812 /* 813 * Conditionally copyin msg.msg_control. 814 */ 815 if (msg.msg_control) { 816 if (msg.msg_controllen < sizeof(struct cmsghdr) || 817 msg.msg_controllen > MLEN) { 818 error = EINVAL; 819 goto cleanup; 820 } 821 control = m_get(MB_WAIT, MT_CONTROL); 822 if (control == NULL) { 823 error = ENOBUFS; 824 goto cleanup; 825 } 826 control->m_len = msg.msg_controllen; 827 error = copyin(msg.msg_control, mtod(control, caddr_t), 828 msg.msg_controllen); 829 if (error) { 830 m_free(control); 831 goto cleanup; 832 } 833 } 834 835 error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags, 836 &uap->sysmsg_szresult); 837 838 cleanup: 839 iovec_free(&iov, aiov); 840 cleanup2: 841 if (sa) 842 kfree(sa, M_SONAME); 843 return (error); 844 } 845 846 /* 847 * kern_recvmsg() takes a handle to sa and control. If the handle is non- 848 * null, it returns a dynamically allocated struct sockaddr and an mbuf. 849 * Don't forget to FREE() and m_free() these if they are returned. 850 */ 851 int 852 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio, 853 struct mbuf **control, int *flags, size_t *res) 854 { 855 struct thread *td = curthread; 856 struct proc *p = td->td_proc; 857 struct file *fp; 858 size_t len; 859 int error; 860 int lflags; 861 struct socket *so; 862 #ifdef KTRACE 863 struct iovec *ktriov = NULL; 864 struct uio ktruio; 865 #endif 866 867 error = holdsock(p->p_fd, s, &fp); 868 if (error) 869 return (error); 870 #ifdef KTRACE 871 if (KTRPOINT(td, KTR_GENIO)) { 872 int iovlen = auio->uio_iovcnt * sizeof (struct iovec); 873 874 ktriov = kmalloc(iovlen, M_TEMP, M_WAITOK); 875 bcopy(auio->uio_iov, ktriov, iovlen); 876 ktruio = *auio; 877 } 878 #endif 879 len = auio->uio_resid; 880 so = (struct socket *)fp->f_data; 881 882 if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) { 883 if (fp->f_flag & FNONBLOCK) { 884 if (flags) { 885 *flags |= MSG_FNONBLOCKING; 886 } else { 887 lflags = MSG_FNONBLOCKING; 888 flags = &lflags; 889 } 890 } 891 } 892 893 error = so_pru_soreceive(so, sa, auio, NULL, control, flags); 894 if (error) { 895 if (auio->uio_resid != len && (error == ERESTART || 896 error == EINTR || error == EWOULDBLOCK)) 897 error = 0; 898 } 899 #ifdef KTRACE 900 if (ktriov != NULL) { 901 if (error == 0) { 902 ktruio.uio_iov = ktriov; 903 ktruio.uio_resid = len - auio->uio_resid; 904 ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error); 905 } 906 kfree(ktriov, M_TEMP); 907 } 908 #endif 909 if (error == 0) 910 *res = len - auio->uio_resid; 911 fdrop(fp); 912 return (error); 913 } 914 915 /* 916 * recvfrom_args(int s, caddr_t buf, size_t len, int flags, 917 * caddr_t from, int *fromlenaddr) 918 * 919 * MPALMOSTSAFE 920 */ 921 int 922 sys_recvfrom(struct recvfrom_args *uap) 923 { 924 struct thread *td = curthread; 925 struct uio auio; 926 struct iovec aiov; 927 struct sockaddr *sa = NULL; 928 int error, fromlen; 929 930 if (uap->from && uap->fromlenaddr) { 931 error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen)); 932 if (error) 933 return (error); 934 if (fromlen < 0) 935 return (EINVAL); 936 } else { 937 fromlen = 0; 938 } 939 aiov.iov_base = uap->buf; 940 aiov.iov_len = uap->len; 941 auio.uio_iov = &aiov; 942 auio.uio_iovcnt = 1; 943 auio.uio_offset = 0; 944 auio.uio_resid = uap->len; 945 auio.uio_segflg = UIO_USERSPACE; 946 auio.uio_rw = UIO_READ; 947 auio.uio_td = td; 948 949 error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL, 950 &uap->flags, &uap->sysmsg_szresult); 951 952 if (error == 0 && uap->from) { 953 /* note: sa may still be NULL */ 954 if (sa) { 955 fromlen = MIN(fromlen, sa->sa_len); 956 error = copyout(sa, uap->from, fromlen); 957 } else { 958 fromlen = 0; 959 } 960 if (error == 0) { 961 error = copyout(&fromlen, uap->fromlenaddr, 962 sizeof(fromlen)); 963 } 964 } 965 if (sa) 966 kfree(sa, M_SONAME); 967 968 return (error); 969 } 970 971 /* 972 * recvmsg_args(int s, struct msghdr *msg, int flags) 973 * 974 * MPALMOSTSAFE 975 */ 976 int 977 sys_recvmsg(struct recvmsg_args *uap) 978 { 979 struct thread *td = curthread; 980 struct msghdr msg; 981 struct uio auio; 982 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 983 struct mbuf *m, *control = NULL; 984 struct sockaddr *sa = NULL; 985 caddr_t ctlbuf; 986 socklen_t *ufromlenp, *ucontrollenp; 987 int error, fromlen, controllen, len, flags, *uflagsp; 988 989 /* 990 * This copyin handles everything except the iovec. 991 */ 992 error = copyin(uap->msg, &msg, sizeof(msg)); 993 if (error) 994 return (error); 995 996 if (msg.msg_name && msg.msg_namelen < 0) 997 return (EINVAL); 998 if (msg.msg_control && msg.msg_controllen < 0) 999 return (EINVAL); 1000 1001 ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1002 msg_namelen)); 1003 ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr, 1004 msg_controllen)); 1005 uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr, 1006 msg_flags)); 1007 1008 /* 1009 * Populate auio. 1010 */ 1011 error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen, 1012 &auio.uio_resid); 1013 if (error) 1014 return (error); 1015 auio.uio_iov = iov; 1016 auio.uio_iovcnt = msg.msg_iovlen; 1017 auio.uio_offset = 0; 1018 auio.uio_segflg = UIO_USERSPACE; 1019 auio.uio_rw = UIO_READ; 1020 auio.uio_td = td; 1021 1022 flags = uap->flags; 1023 1024 error = kern_recvmsg(uap->s, 1025 (msg.msg_name ? &sa : NULL), &auio, 1026 (msg.msg_control ? &control : NULL), &flags, 1027 &uap->sysmsg_szresult); 1028 1029 /* 1030 * Conditionally copyout the name and populate the namelen field. 1031 */ 1032 if (error == 0 && msg.msg_name) { 1033 /* note: sa may still be NULL */ 1034 if (sa != NULL) { 1035 fromlen = MIN(msg.msg_namelen, sa->sa_len); 1036 error = copyout(sa, msg.msg_name, fromlen); 1037 } else { 1038 fromlen = 0; 1039 } 1040 if (error == 0) 1041 error = copyout(&fromlen, ufromlenp, 1042 sizeof(*ufromlenp)); 1043 } 1044 1045 /* 1046 * Copyout msg.msg_control and msg.msg_controllen. 1047 */ 1048 if (error == 0 && msg.msg_control) { 1049 len = msg.msg_controllen; 1050 m = control; 1051 ctlbuf = (caddr_t)msg.msg_control; 1052 1053 while(m && len > 0) { 1054 unsigned int tocopy; 1055 1056 if (len >= m->m_len) { 1057 tocopy = m->m_len; 1058 } else { 1059 msg.msg_flags |= MSG_CTRUNC; 1060 tocopy = len; 1061 } 1062 1063 error = copyout(mtod(m, caddr_t), ctlbuf, tocopy); 1064 if (error) 1065 goto cleanup; 1066 1067 ctlbuf += tocopy; 1068 len -= tocopy; 1069 m = m->m_next; 1070 } 1071 controllen = ctlbuf - (caddr_t)msg.msg_control; 1072 error = copyout(&controllen, ucontrollenp, 1073 sizeof(*ucontrollenp)); 1074 } 1075 1076 if (error == 0) 1077 error = copyout(&flags, uflagsp, sizeof(*uflagsp)); 1078 1079 cleanup: 1080 if (sa) 1081 kfree(sa, M_SONAME); 1082 iovec_free(&iov, aiov); 1083 if (control) 1084 m_freem(control); 1085 return (error); 1086 } 1087 1088 /* 1089 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1090 * in kernel pointer instead of a userland pointer. This allows us 1091 * to manipulate socket options in the emulation code. 1092 */ 1093 int 1094 kern_setsockopt(int s, struct sockopt *sopt) 1095 { 1096 struct thread *td = curthread; 1097 struct proc *p = td->td_proc; 1098 struct file *fp; 1099 int error; 1100 1101 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1102 return (EFAULT); 1103 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1104 return (EINVAL); 1105 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1106 return (EINVAL); 1107 1108 error = holdsock(p->p_fd, s, &fp); 1109 if (error) 1110 return (error); 1111 1112 error = sosetopt((struct socket *)fp->f_data, sopt); 1113 fdrop(fp); 1114 return (error); 1115 } 1116 1117 /* 1118 * setsockopt_args(int s, int level, int name, caddr_t val, int valsize) 1119 * 1120 * MPALMOSTSAFE 1121 */ 1122 int 1123 sys_setsockopt(struct setsockopt_args *uap) 1124 { 1125 struct thread *td = curthread; 1126 struct sockopt sopt; 1127 int error; 1128 1129 sopt.sopt_level = uap->level; 1130 sopt.sopt_name = uap->name; 1131 sopt.sopt_valsize = uap->valsize; 1132 sopt.sopt_td = td; 1133 sopt.sopt_val = NULL; 1134 1135 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1136 return (EINVAL); 1137 if (uap->val) { 1138 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1139 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1140 if (error) 1141 goto out; 1142 } 1143 1144 error = kern_setsockopt(uap->s, &sopt); 1145 out: 1146 if (uap->val) 1147 kfree(sopt.sopt_val, M_TEMP); 1148 return(error); 1149 } 1150 1151 /* 1152 * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an 1153 * in kernel pointer instead of a userland pointer. This allows us 1154 * to manipulate socket options in the emulation code. 1155 */ 1156 int 1157 kern_getsockopt(int s, struct sockopt *sopt) 1158 { 1159 struct thread *td = curthread; 1160 struct proc *p = td->td_proc; 1161 struct file *fp; 1162 int error; 1163 1164 if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0) 1165 return (EFAULT); 1166 if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0) 1167 return (EINVAL); 1168 if (sopt->sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1169 return (EINVAL); 1170 1171 error = holdsock(p->p_fd, s, &fp); 1172 if (error) 1173 return (error); 1174 1175 error = sogetopt((struct socket *)fp->f_data, sopt); 1176 fdrop(fp); 1177 return (error); 1178 } 1179 1180 /* 1181 * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize) 1182 * 1183 * MPALMOSTSAFE 1184 */ 1185 int 1186 sys_getsockopt(struct getsockopt_args *uap) 1187 { 1188 struct thread *td = curthread; 1189 struct sockopt sopt; 1190 int error, valsize; 1191 1192 if (uap->val) { 1193 error = copyin(uap->avalsize, &valsize, sizeof(valsize)); 1194 if (error) 1195 return (error); 1196 } else { 1197 valsize = 0; 1198 } 1199 1200 sopt.sopt_level = uap->level; 1201 sopt.sopt_name = uap->name; 1202 sopt.sopt_valsize = valsize; 1203 sopt.sopt_td = td; 1204 sopt.sopt_val = NULL; 1205 1206 if (sopt.sopt_valsize > SOMAXOPT_SIZE) /* unsigned */ 1207 return (EINVAL); 1208 if (uap->val) { 1209 sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK); 1210 error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize); 1211 if (error) 1212 goto out; 1213 } 1214 1215 error = kern_getsockopt(uap->s, &sopt); 1216 if (error) 1217 goto out; 1218 valsize = sopt.sopt_valsize; 1219 error = copyout(&valsize, uap->avalsize, sizeof(valsize)); 1220 if (error) 1221 goto out; 1222 if (uap->val) 1223 error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize); 1224 out: 1225 if (uap->val) 1226 kfree(sopt.sopt_val, M_TEMP); 1227 return (error); 1228 } 1229 1230 /* 1231 * The second argument to kern_getsockname() is a handle to a struct sockaddr. 1232 * This allows kern_getsockname() to return a pointer to an allocated struct 1233 * sockaddr which must be freed later with FREE(). The caller must 1234 * initialize *name to NULL. 1235 */ 1236 int 1237 kern_getsockname(int s, struct sockaddr **name, int *namelen) 1238 { 1239 struct thread *td = curthread; 1240 struct proc *p = td->td_proc; 1241 struct file *fp; 1242 struct socket *so; 1243 struct sockaddr *sa = NULL; 1244 int error; 1245 1246 error = holdsock(p->p_fd, s, &fp); 1247 if (error) 1248 return (error); 1249 if (*namelen < 0) { 1250 fdrop(fp); 1251 return (EINVAL); 1252 } 1253 so = (struct socket *)fp->f_data; 1254 error = so_pru_sockaddr(so, &sa); 1255 if (error == 0) { 1256 if (sa == NULL) { 1257 *namelen = 0; 1258 } else { 1259 *namelen = MIN(*namelen, sa->sa_len); 1260 *name = sa; 1261 } 1262 } 1263 1264 fdrop(fp); 1265 return (error); 1266 } 1267 1268 /* 1269 * getsockname_args(int fdes, caddr_t asa, int *alen) 1270 * 1271 * Get socket name. 1272 * 1273 * MPALMOSTSAFE 1274 */ 1275 int 1276 sys_getsockname(struct getsockname_args *uap) 1277 { 1278 struct sockaddr *sa = NULL; 1279 int error, sa_len; 1280 1281 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1282 if (error) 1283 return (error); 1284 1285 error = kern_getsockname(uap->fdes, &sa, &sa_len); 1286 1287 if (error == 0) 1288 error = copyout(sa, uap->asa, sa_len); 1289 if (error == 0) 1290 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1291 if (sa) 1292 kfree(sa, M_SONAME); 1293 return (error); 1294 } 1295 1296 /* 1297 * The second argument to kern_getpeername() is a handle to a struct sockaddr. 1298 * This allows kern_getpeername() to return a pointer to an allocated struct 1299 * sockaddr which must be freed later with FREE(). The caller must 1300 * initialize *name to NULL. 1301 */ 1302 int 1303 kern_getpeername(int s, struct sockaddr **name, int *namelen) 1304 { 1305 struct thread *td = curthread; 1306 struct proc *p = td->td_proc; 1307 struct file *fp; 1308 struct socket *so; 1309 struct sockaddr *sa = NULL; 1310 int error; 1311 1312 error = holdsock(p->p_fd, s, &fp); 1313 if (error) 1314 return (error); 1315 if (*namelen < 0) { 1316 fdrop(fp); 1317 return (EINVAL); 1318 } 1319 so = (struct socket *)fp->f_data; 1320 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) { 1321 fdrop(fp); 1322 return (ENOTCONN); 1323 } 1324 error = so_pru_peeraddr(so, &sa); 1325 if (error == 0) { 1326 if (sa == NULL) { 1327 *namelen = 0; 1328 } else { 1329 *namelen = MIN(*namelen, sa->sa_len); 1330 *name = sa; 1331 } 1332 } 1333 1334 fdrop(fp); 1335 return (error); 1336 } 1337 1338 /* 1339 * getpeername_args(int fdes, caddr_t asa, int *alen) 1340 * 1341 * Get name of peer for connected socket. 1342 * 1343 * MPALMOSTSAFE 1344 */ 1345 int 1346 sys_getpeername(struct getpeername_args *uap) 1347 { 1348 struct sockaddr *sa = NULL; 1349 int error, sa_len; 1350 1351 error = copyin(uap->alen, &sa_len, sizeof(sa_len)); 1352 if (error) 1353 return (error); 1354 1355 error = kern_getpeername(uap->fdes, &sa, &sa_len); 1356 1357 if (error == 0) 1358 error = copyout(sa, uap->asa, sa_len); 1359 if (error == 0) 1360 error = copyout(&sa_len, uap->alen, sizeof(*uap->alen)); 1361 if (sa) 1362 kfree(sa, M_SONAME); 1363 return (error); 1364 } 1365 1366 int 1367 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len) 1368 { 1369 struct sockaddr *sa; 1370 int error; 1371 1372 *namp = NULL; 1373 if (len > SOCK_MAXADDRLEN) 1374 return ENAMETOOLONG; 1375 if (len < offsetof(struct sockaddr, sa_data[0])) 1376 return EDOM; 1377 sa = kmalloc(len, M_SONAME, M_WAITOK); 1378 error = copyin(uaddr, sa, len); 1379 if (error) { 1380 kfree(sa, M_SONAME); 1381 } else { 1382 #if BYTE_ORDER != BIG_ENDIAN 1383 /* 1384 * The bind(), connect(), and sendto() syscalls were not 1385 * versioned for COMPAT_43. Thus, this check must stay. 1386 */ 1387 if (sa->sa_family == 0 && sa->sa_len < AF_MAX) 1388 sa->sa_family = sa->sa_len; 1389 #endif 1390 sa->sa_len = len; 1391 *namp = sa; 1392 } 1393 return error; 1394 } 1395 1396 /* 1397 * Detach a mapped page and release resources back to the system. 1398 * We must release our wiring and if the object is ripped out 1399 * from under the vm_page we become responsible for freeing the 1400 * page. 1401 * 1402 * MPSAFE 1403 */ 1404 static void 1405 sf_buf_mfree(void *arg) 1406 { 1407 struct sf_buf *sf = arg; 1408 vm_page_t m; 1409 1410 m = sf_buf_page(sf); 1411 if (sf_buf_free(sf)) { 1412 /* sf invalid now */ 1413 /* 1414 vm_page_busy_wait(m, FALSE, "sockpgf"); 1415 vm_page_wakeup(m); 1416 */ 1417 vm_page_unhold(m); 1418 #if 0 1419 if (m->object == NULL && 1420 m->wire_count == 0 && 1421 (m->flags & PG_NEED_COMMIT) == 0) { 1422 vm_page_free(m); 1423 } else { 1424 vm_page_wakeup(m); 1425 } 1426 #endif 1427 } 1428 } 1429 1430 /* 1431 * sendfile(2). 1432 * int sendfile(int fd, int s, off_t offset, size_t nbytes, 1433 * struct sf_hdtr *hdtr, off_t *sbytes, int flags) 1434 * 1435 * Send a file specified by 'fd' and starting at 'offset' to a socket 1436 * specified by 's'. Send only 'nbytes' of the file or until EOF if 1437 * nbytes == 0. Optionally add a header and/or trailer to the socket 1438 * output. If specified, write the total number of bytes sent into *sbytes. 1439 * 1440 * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused 1441 * the headers to count against the remaining bytes to be sent from 1442 * the file descriptor. We may wish to implement a compatibility syscall 1443 * in the future. 1444 * 1445 * MPALMOSTSAFE 1446 */ 1447 int 1448 sys_sendfile(struct sendfile_args *uap) 1449 { 1450 struct thread *td = curthread; 1451 struct proc *p = td->td_proc; 1452 struct file *fp; 1453 struct vnode *vp = NULL; 1454 struct sf_hdtr hdtr; 1455 struct iovec aiov[UIO_SMALLIOV], *iov = NULL; 1456 struct uio auio; 1457 struct mbuf *mheader = NULL; 1458 size_t hbytes = 0; 1459 size_t tbytes; 1460 off_t hdtr_size = 0; 1461 off_t sbytes; 1462 int error; 1463 1464 KKASSERT(p); 1465 1466 /* 1467 * Do argument checking. Must be a regular file in, stream 1468 * type and connected socket out, positive offset. 1469 */ 1470 fp = holdfp(p->p_fd, uap->fd, FREAD); 1471 if (fp == NULL) { 1472 return (EBADF); 1473 } 1474 if (fp->f_type != DTYPE_VNODE) { 1475 fdrop(fp); 1476 return (EINVAL); 1477 } 1478 vp = (struct vnode *)fp->f_data; 1479 vref(vp); 1480 fdrop(fp); 1481 1482 /* 1483 * If specified, get the pointer to the sf_hdtr struct for 1484 * any headers/trailers. 1485 */ 1486 if (uap->hdtr) { 1487 error = copyin(uap->hdtr, &hdtr, sizeof(hdtr)); 1488 if (error) 1489 goto done; 1490 /* 1491 * Send any headers. 1492 */ 1493 if (hdtr.headers) { 1494 error = iovec_copyin(hdtr.headers, &iov, aiov, 1495 hdtr.hdr_cnt, &hbytes); 1496 if (error) 1497 goto done; 1498 auio.uio_iov = iov; 1499 auio.uio_iovcnt = hdtr.hdr_cnt; 1500 auio.uio_offset = 0; 1501 auio.uio_segflg = UIO_USERSPACE; 1502 auio.uio_rw = UIO_WRITE; 1503 auio.uio_td = td; 1504 auio.uio_resid = hbytes; 1505 1506 mheader = m_uiomove(&auio); 1507 1508 iovec_free(&iov, aiov); 1509 if (mheader == NULL) 1510 goto done; 1511 } 1512 } 1513 1514 error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader, 1515 &sbytes, uap->flags); 1516 if (error) 1517 goto done; 1518 1519 /* 1520 * Send trailers. Wimp out and use writev(2). 1521 */ 1522 if (uap->hdtr != NULL && hdtr.trailers != NULL) { 1523 error = iovec_copyin(hdtr.trailers, &iov, aiov, 1524 hdtr.trl_cnt, &auio.uio_resid); 1525 if (error) 1526 goto done; 1527 auio.uio_iov = iov; 1528 auio.uio_iovcnt = hdtr.trl_cnt; 1529 auio.uio_offset = 0; 1530 auio.uio_segflg = UIO_USERSPACE; 1531 auio.uio_rw = UIO_WRITE; 1532 auio.uio_td = td; 1533 1534 error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes); 1535 1536 iovec_free(&iov, aiov); 1537 if (error) 1538 goto done; 1539 hdtr_size += tbytes; /* trailer bytes successfully sent */ 1540 } 1541 1542 done: 1543 if (vp) 1544 vrele(vp); 1545 if (uap->sbytes != NULL) { 1546 sbytes += hdtr_size; 1547 copyout(&sbytes, uap->sbytes, sizeof(off_t)); 1548 } 1549 return (error); 1550 } 1551 1552 int 1553 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes, 1554 struct mbuf *mheader, off_t *sbytes, int flags) 1555 { 1556 struct thread *td = curthread; 1557 struct proc *p = td->td_proc; 1558 struct vm_object *obj; 1559 struct socket *so; 1560 struct file *fp; 1561 struct mbuf *m, *mp; 1562 struct sf_buf *sf; 1563 struct vm_page *pg; 1564 off_t off, xfsize, xbytes; 1565 off_t hbytes = 0; 1566 int error = 0; 1567 1568 if (vp->v_type != VREG) { 1569 error = EINVAL; 1570 goto done0; 1571 } 1572 if ((obj = vp->v_object) == NULL) { 1573 error = EINVAL; 1574 goto done0; 1575 } 1576 error = holdsock(p->p_fd, sfd, &fp); 1577 if (error) 1578 goto done0; 1579 so = (struct socket *)fp->f_data; 1580 if (so->so_type != SOCK_STREAM) { 1581 error = EINVAL; 1582 goto done; 1583 } 1584 if ((so->so_state & SS_ISCONNECTED) == 0) { 1585 error = ENOTCONN; 1586 goto done; 1587 } 1588 if (offset < 0) { 1589 error = EINVAL; 1590 goto done; 1591 } 1592 1593 /* 1594 * preallocation is required for asynchronous passing of mbufs, 1595 * otherwise we can wind up building up an infinite number of 1596 * mbufs during the asynchronous latency. 1597 */ 1598 if ((so->so_snd.ssb_flags & (SSB_PREALLOC | SSB_STOPSUPP)) == 0) { 1599 error = EINVAL; 1600 goto done; 1601 } 1602 1603 *sbytes = 0; 1604 xbytes = 0; 1605 /* 1606 * Protect against multiple writers to the socket. 1607 */ 1608 ssb_lock(&so->so_snd, M_WAITOK); 1609 1610 /* 1611 * Loop through the pages in the file, starting with the requested 1612 * offset. Get a file page (do I/O if necessary), map the file page 1613 * into an sf_buf, attach an mbuf header to the sf_buf, and queue 1614 * it on the socket. 1615 */ 1616 for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes, xbytes += xfsize) { 1617 vm_pindex_t pindex; 1618 vm_offset_t pgoff; 1619 long space; 1620 1621 pindex = OFF_TO_IDX(off); 1622 retry_lookup: 1623 /* 1624 * Calculate the amount to transfer. Not to exceed a page, 1625 * the EOF, or the passed in nbytes. 1626 */ 1627 xfsize = vp->v_filesize - off; 1628 if (xfsize > PAGE_SIZE) 1629 xfsize = PAGE_SIZE; 1630 pgoff = (vm_offset_t)(off & PAGE_MASK); 1631 if (PAGE_SIZE - pgoff < xfsize) 1632 xfsize = PAGE_SIZE - pgoff; 1633 if (nbytes && xfsize > (nbytes - xbytes)) 1634 xfsize = nbytes - xbytes; 1635 if (xfsize <= 0) 1636 break; 1637 /* 1638 * Optimize the non-blocking case by looking at the socket space 1639 * before going to the extra work of constituting the sf_buf. 1640 */ 1641 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1642 space = ssb_space_prealloc(&so->so_snd); 1643 else 1644 space = ssb_space(&so->so_snd); 1645 1646 if ((fp->f_flag & FNONBLOCK) && space <= 0) { 1647 if (so->so_state & SS_CANTSENDMORE) 1648 error = EPIPE; 1649 else 1650 error = EAGAIN; 1651 ssb_unlock(&so->so_snd); 1652 goto done; 1653 } 1654 /* 1655 * Attempt to look up the page. 1656 * 1657 * Allocate if not found, wait and loop if busy, then hold the page. 1658 * We hold rather than wire the page because we do not want to prevent 1659 * filesystem truncation operations from occuring on the file. This 1660 * can happen even under normal operation if the file being sent is 1661 * remove()d after the sendfile() call completes, because the socket buffer 1662 * may still be draining. tmpfs will crash if we try to use wire. 1663 */ 1664 vm_object_hold(obj); 1665 pg = vm_page_lookup_busy_try(obj, pindex, TRUE, &error); 1666 if (error) { 1667 vm_page_sleep_busy(pg, TRUE, "sfpbsy"); 1668 vm_object_drop(obj); 1669 goto retry_lookup; 1670 } 1671 if (pg == NULL) { 1672 pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL | 1673 VM_ALLOC_NULL_OK); 1674 if (pg == NULL) { 1675 vm_wait(0); 1676 vm_object_drop(obj); 1677 goto retry_lookup; 1678 } 1679 } 1680 vm_page_hold(pg); 1681 vm_object_drop(obj); 1682 1683 /* 1684 * If page is not valid for what we need, initiate I/O 1685 */ 1686 1687 if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) { 1688 struct uio auio; 1689 struct iovec aiov; 1690 int bsize; 1691 1692 /* 1693 * Ensure that our page is still around when the I/O 1694 * completes. 1695 * 1696 * Ensure that our page is not modified while part of 1697 * a mbuf as this could mess up tcp checksums, DMA, 1698 * etc (XXX NEEDS WORK). The softbusy is supposed to 1699 * help here but it actually doesn't. 1700 * 1701 * XXX THIS HAS MULTIPLE PROBLEMS. The underlying 1702 * VM pages are not protected by the soft-busy 1703 * unless we vm_page_protect... READ them, and 1704 * they STILL aren't protected against 1705 * modification via the buffer cache (VOP_WRITE). 1706 * 1707 * Fixing the second issue is particularly 1708 * difficult. 1709 * 1710 * XXX We also can't soft-busy anyway because it can 1711 * deadlock against the syncer doing a vfs_msync(), 1712 * vfs_msync->vmntvnodesca->vfs_msync_scan2-> 1713 * vm_object_page_clean->(scan)-> ... page 1714 * busy-wait. 1715 */ 1716 /*vm_page_io_start(pg);*/ 1717 vm_page_wakeup(pg); 1718 1719 /* 1720 * Get the page from backing store. 1721 */ 1722 bsize = vp->v_mount->mnt_stat.f_iosize; 1723 auio.uio_iov = &aiov; 1724 auio.uio_iovcnt = 1; 1725 aiov.iov_base = 0; 1726 aiov.iov_len = MAXBSIZE; 1727 auio.uio_resid = MAXBSIZE; 1728 auio.uio_offset = trunc_page(off); 1729 auio.uio_segflg = UIO_NOCOPY; 1730 auio.uio_rw = UIO_READ; 1731 auio.uio_td = td; 1732 vn_lock(vp, LK_SHARED | LK_RETRY); 1733 error = VOP_READ(vp, &auio, 1734 IO_VMIO | ((MAXBSIZE / bsize) << 16), 1735 td->td_ucred); 1736 vn_unlock(vp); 1737 vm_page_flag_clear(pg, PG_ZERO); 1738 vm_page_busy_wait(pg, FALSE, "sockpg"); 1739 /*vm_page_io_finish(pg);*/ 1740 if (error) { 1741 vm_page_wakeup(pg); 1742 vm_page_unhold(pg); 1743 /* vm_page_try_to_free(pg); */ 1744 ssb_unlock(&so->so_snd); 1745 goto done; 1746 } 1747 } 1748 1749 1750 /* 1751 * Get a sendfile buf. We usually wait as long as necessary, 1752 * but this wait can be interrupted. 1753 */ 1754 if ((sf = sf_buf_alloc(pg)) == NULL) { 1755 vm_page_wakeup(pg); 1756 vm_page_unhold(pg); 1757 /* vm_page_try_to_free(pg); */ 1758 ssb_unlock(&so->so_snd); 1759 error = EINTR; 1760 goto done; 1761 } 1762 1763 /* 1764 * Get an mbuf header and set it up as having external storage. 1765 */ 1766 MGETHDR(m, MB_WAIT, MT_DATA); 1767 if (m == NULL) { 1768 error = ENOBUFS; 1769 vm_page_wakeup(pg); 1770 vm_page_unhold(pg); 1771 /* vm_page_try_to_free(pg); */ 1772 sf_buf_free(sf); 1773 ssb_unlock(&so->so_snd); 1774 goto done; 1775 } 1776 1777 vm_page_wakeup(pg); 1778 1779 m->m_ext.ext_free = sf_buf_mfree; 1780 m->m_ext.ext_ref = sf_buf_ref; 1781 m->m_ext.ext_arg = sf; 1782 m->m_ext.ext_buf = (void *)sf_buf_kva(sf); 1783 m->m_ext.ext_size = PAGE_SIZE; 1784 m->m_data = (char *)sf_buf_kva(sf) + pgoff; 1785 m->m_flags |= M_EXT; 1786 m->m_pkthdr.len = m->m_len = xfsize; 1787 KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0); 1788 1789 if (mheader != NULL) { 1790 hbytes = mheader->m_pkthdr.len; 1791 mheader->m_pkthdr.len += m->m_pkthdr.len; 1792 m_cat(mheader, m); 1793 m = mheader; 1794 mheader = NULL; 1795 } else 1796 hbytes = 0; 1797 1798 /* 1799 * Add the buffer to the socket buffer chain. 1800 */ 1801 crit_enter(); 1802 retry_space: 1803 /* 1804 * Make sure that the socket is still able to take more data. 1805 * CANTSENDMORE being true usually means that the connection 1806 * was closed. so_error is true when an error was sensed after 1807 * a previous send. 1808 * The state is checked after the page mapping and buffer 1809 * allocation above since those operations may block and make 1810 * any socket checks stale. From this point forward, nothing 1811 * blocks before the pru_send (or more accurately, any blocking 1812 * results in a loop back to here to re-check). 1813 */ 1814 if ((so->so_state & SS_CANTSENDMORE) || so->so_error) { 1815 if (so->so_state & SS_CANTSENDMORE) { 1816 error = EPIPE; 1817 } else { 1818 error = so->so_error; 1819 so->so_error = 0; 1820 } 1821 m_freem(m); 1822 ssb_unlock(&so->so_snd); 1823 crit_exit(); 1824 goto done; 1825 } 1826 /* 1827 * Wait for socket space to become available. We do this just 1828 * after checking the connection state above in order to avoid 1829 * a race condition with ssb_wait(). 1830 */ 1831 if (so->so_snd.ssb_flags & SSB_PREALLOC) 1832 space = ssb_space_prealloc(&so->so_snd); 1833 else 1834 space = ssb_space(&so->so_snd); 1835 1836 if (space < m->m_pkthdr.len && space < so->so_snd.ssb_lowat) { 1837 if (fp->f_flag & FNONBLOCK) { 1838 m_freem(m); 1839 ssb_unlock(&so->so_snd); 1840 crit_exit(); 1841 error = EAGAIN; 1842 goto done; 1843 } 1844 error = ssb_wait(&so->so_snd); 1845 /* 1846 * An error from ssb_wait usually indicates that we've 1847 * been interrupted by a signal. If we've sent anything 1848 * then return bytes sent, otherwise return the error. 1849 */ 1850 if (error) { 1851 m_freem(m); 1852 ssb_unlock(&so->so_snd); 1853 crit_exit(); 1854 goto done; 1855 } 1856 goto retry_space; 1857 } 1858 1859 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1860 for (mp = m; mp != NULL; mp = mp->m_next) 1861 ssb_preallocstream(&so->so_snd, mp); 1862 } 1863 if (use_sendfile_async) 1864 error = so_pru_senda(so, 0, m, NULL, NULL, td); 1865 else 1866 error = so_pru_send(so, 0, m, NULL, NULL, td); 1867 1868 crit_exit(); 1869 if (error) { 1870 ssb_unlock(&so->so_snd); 1871 goto done; 1872 } 1873 } 1874 if (mheader != NULL) { 1875 *sbytes += mheader->m_pkthdr.len; 1876 1877 if (so->so_snd.ssb_flags & SSB_PREALLOC) { 1878 for (mp = mheader; mp != NULL; mp = mp->m_next) 1879 ssb_preallocstream(&so->so_snd, mp); 1880 } 1881 if (use_sendfile_async) 1882 error = so_pru_senda(so, 0, mheader, NULL, NULL, td); 1883 else 1884 error = so_pru_send(so, 0, mheader, NULL, NULL, td); 1885 1886 mheader = NULL; 1887 } 1888 ssb_unlock(&so->so_snd); 1889 1890 done: 1891 fdrop(fp); 1892 done0: 1893 if (mheader != NULL) 1894 m_freem(mheader); 1895 return (error); 1896 } 1897 1898 /* 1899 * MPALMOSTSAFE 1900 */ 1901 int 1902 sys_sctp_peeloff(struct sctp_peeloff_args *uap) 1903 { 1904 #ifdef SCTP 1905 struct thread *td = curthread; 1906 struct filedesc *fdp = td->td_proc->p_fd; 1907 struct file *lfp = NULL; 1908 struct file *nfp = NULL; 1909 int error; 1910 struct socket *head, *so; 1911 caddr_t assoc_id; 1912 int fd; 1913 short fflag; /* type must match fp->f_flag */ 1914 1915 assoc_id = uap->name; 1916 error = holdsock(td->td_proc->p_fd, uap->sd, &lfp); 1917 if (error) 1918 return (error); 1919 1920 crit_enter(); 1921 head = (struct socket *)lfp->f_data; 1922 error = sctp_can_peel_off(head, assoc_id); 1923 if (error) { 1924 crit_exit(); 1925 goto done; 1926 } 1927 /* 1928 * At this point we know we do have a assoc to pull 1929 * we proceed to get the fd setup. This may block 1930 * but that is ok. 1931 */ 1932 1933 fflag = lfp->f_flag; 1934 error = falloc(td->td_lwp, &nfp, &fd); 1935 if (error) { 1936 /* 1937 * Probably ran out of file descriptors. Put the 1938 * unaccepted connection back onto the queue and 1939 * do another wakeup so some other process might 1940 * have a chance at it. 1941 */ 1942 crit_exit(); 1943 goto done; 1944 } 1945 uap->sysmsg_iresult = fd; 1946 1947 so = sctp_get_peeloff(head, assoc_id, &error); 1948 if (so == NULL) { 1949 /* 1950 * Either someone else peeled it off OR 1951 * we can't get a socket. 1952 */ 1953 goto noconnection; 1954 } 1955 soreference(so); /* reference needed */ 1956 soclrstate(so, SS_NOFDREF | SS_COMP); /* when clearing NOFDREF */ 1957 so->so_head = NULL; 1958 if (head->so_sigio != NULL) 1959 fsetown(fgetown(&head->so_sigio), &so->so_sigio); 1960 1961 nfp->f_type = DTYPE_SOCKET; 1962 nfp->f_flag = fflag; 1963 nfp->f_ops = &socketops; 1964 nfp->f_data = so; 1965 1966 noconnection: 1967 /* 1968 * Assign the file pointer to the reserved descriptor, or clear 1969 * the reserved descriptor if an error occured. 1970 */ 1971 if (error) 1972 fsetfd(fdp, NULL, fd); 1973 else 1974 fsetfd(fdp, nfp, fd); 1975 crit_exit(); 1976 /* 1977 * Release explicitly held references before returning. 1978 */ 1979 done: 1980 if (nfp != NULL) 1981 fdrop(nfp); 1982 fdrop(lfp); 1983 return (error); 1984 #else /* SCTP */ 1985 return(EOPNOTSUPP); 1986 #endif /* SCTP */ 1987 } 1988